Debt covenant slack and real earnings management

Debt covenant slack and real earnings management Bong Hwan Kim American University Kogod School of Business 4400 Massachusetts Avenue, NW Washington, DC 20016 bkim@american.edu Ling Lei George Mason University School of Management 4400 University Drive MS 5F4 Fairfax, VA 22030 llei1@gmu.edu Mikhail Pevzner* George Mason University School of Management 4400 University Drive MS 5F4 Fairfax, VA 22030 mpevzner@gmu.edu November 2010 *Corresponding author. We thank J.K Aier, Dan Cohen, Rich Frankel, Keith Jones, Karen Kitching, Dino Silveri, workshop participants at George Mason University for their valuable comments, and Michael Roberts for providing us with the Dealscan- Compustat matching dataset. Kim thanks Washington University s Olin School of Business doctoral research program which provided financial and research support in the initial stages of this paper s development. Pevzner thanks GMU Provost office and GMU Accounting Area Excellence Fund for supporting his summer 2010 research activities.

Debt covenant slack and real earnings management Abstract We examine the relation between firms real earnings management decisions and the slack in their net worth debt covenant. Using private debt covenant data, we find that the overall level of real earnings management is higher when net worth covenant slack is tighter. Moreover, we find that this effect is more pronounced for loan-years with the tightest slack, which is a setting where benefits of managing earnings are greater. Within the sub-sample of loan-years with the tightest slack, we find that real earnings management is higher for borrowers that experienced increases in bankruptcy risk in the previous year. These results suggest that: (1) firms use real earnings management to avoid violations of debt covenants; and (2) that they are more likely to do so when their ability to renegotiate the technical covenant violations is restricted. Our results are largely robust to controlling for endogeneity of the tightness of debt covenant slack. Finally, we find that the positive relation between the tightness of debt covenant slack and real earnings management exists both before and after the adoption of the Sarbanes-Oxley Act. Keywords: debt covenants, debt covenant slack, real earnings management, Dealscan JEL Classification: M40, M41 Data availability: all data is available from commercial sources identified in the text.

1. Introduction We examine whether real earnings management is more pronounced when firms get closer to the violations of their net worth debt covenants, i.e. when firms have tighter debt covenant slack. We further examine whether this relation is more pronounced for firms with the tightest slack, and for firms that have experienced increases in bankruptcy risk, i.e. firms that are expected to have greater difficulties renegotiating their debt contracts and as a result incur a higher cost of covenant violations. We also test whether the relation between the tightness of debt covenant slack and real earnings management changes after the Sarbanes-Oxley Act (SOX). Positive Accounting Theory s Debt Covenant Hypothesis predicts that firms with tighter slack are more likely to manage earnings in order to avoid possible future debt covenant violations (Watts and Zimmerman, 1986). Debt covenant slack measures the level of a firm s proximity to debt covenant violation. The tighter the amount of slack, the greater the likelihood that a borrower would violate a debt covenant. Prior research shows that disclosed debt covenant violations are costly because such violations increase a firm s cost of debt capital and lead to the reduction of potentially useful investments (Chava and Roberts, 2008, Roberts and Sufi, 2009a). 1 Often firms end up only in technical default, i.e. they violate debt covenants but manage to avoid the need to disclose these violations through re-negotiation with the lender or by obtaining a 1 Technically, only those violations which are not cured as of the financial statement date have to be reported in SEC filings. On that basis, Dichev and Skinner (2002) and Armstrong et al. (2009) note that disclosed debt covenant violations represent a small sub-sample of debt covenants which could not be renegotiated or whose violations were not waved. Those are most likely to be firms in the most severe financial distress. However, Roberts and Sufi (2009a) who compare the violations disclosed in SEC filings with those reported in Dealscan, show that often firms voluntarily disclose their cured covenant violations. The potential implication of Roberts and Sufi s results is that even cured covenant violations result in a significant negative impact on firms cost of debt. 2

covenant waiver from a lender. Because technical defaults are common 2, one could argue that borrowers should reasonably expect to renegotiate covenant violations at low cost, thereby removing the need to take opportunistic actions to avoid debt covenant violations. However, Chava and Roberts (2008) show that technical defaults, despite the available option of renegotiation, are associated with the transfer of control rights from shareholders to lenders, which leads to a reduction in firms capital expenditures. In addition, prior research shows that renegotiations after technical defaults are associated with less favorable renegotiated contract terms. Roberts and Sufi (2009b) report that private debt covenant violations or technical defaults are significant predictors of unfavorable outcomes of future loan renegotiations, e.g., reductions in loan principal or increases in loan spread. Furthermore, the renegotiation process is likely to be costly because it involves other valuable resources such as the use of lawyers, auditors, firm managers and accounting personnel. Finally, even if a borrower obtains a covenant violation waiver or renegotiates a debt agreement, as long as any renegotiated amendments to the loan contract are material, such renegotiations will have to be disclosed in a form 8-K. The market could react adversely to such disclosures if it perceives them as negative indicators of future performance. Hence, avoiding debt covenant violations, whether disclosed or not, is a powerful incentive to manage earnings (Watts and Zimmerman, 1986). Consistent with this, prior research suggests that firms manage accruals and become less conservative in order to avoid debt covenant violations (Defond and Jiambalvo, 1994, Dichev and Skinner, 2002, Zhang, 2008, Kim, 2009). However, accrual earnings 2 Dichev and Skinner (2002) document an incidence of 30% in Dealscan. See also Chava and Roberts (2008). 3

management to avoid covenant violations is potentially more costly to managers than real earnings management, because accruals earnings management is more likely to draw auditor and regulatory scrutiny (Roychowdhury 2006, Cohen et al. 2008, Zhang, 2007, Cohen and Zarowin, 2010). Thus, managers may prefer real earnings management to accruals earnings management to avoid debt covenant violations. 3 Bartov (1993) and Haw et al. (1991) find initial supporting evidence that firms use real activities manipulation to avoid debt covenant violations through asset sales or through settling of over-funded pension plans. We further extend these initial findings by: 1) directly examining private debt contracts and avoiding the need to rely on the leverage ratio as an indirect proxy of debt covenant restrictiveness; 2) showing that even with the possibility of covenant violation renegotiations, borrowers still prefer real earnings management to renegotiation, particularly when renegotiation costs are likely to be higher, e.g., when borrowers experience increases in bankruptcy risk; 4 and 3) examining real activities in more general terms than asset sales or settling of over-funded pension plans. Following Roychowdhury (2006) and Cohen et al. (2008), we proxy for real earnings management using estimates of a firm s abnormal cash flows, abnormal inventory production, abnormal discretionary expenditures, and a summary measure combining all three of these components. Following Chava and Roberts (2008), we focus on net worth and tangible net worth covenants in our analysis because such covenants are frequently 3 Cohen and Zarowin (2010) cite survey evidence in Graham et al. (2005) documenting that managers are willing to resort to real earnings management even when it leads to the avoidance of valuable projects for a firm. 4 This is important because Haw et al. (1991) argue that earnings management should be less likely to occur in private debt contracts due to the perceived ability to renegotiate. For that reason, Haw et al. (1991) focus on public debt. 4

reported in the Dealscan database, and the measurement of slack in those covenant types is unambiguous. 5 In addition, the effect of earnings management on net worth covenants is clearer because net worth is directly affected by income. 6 Also consistent with Chava and Roberts (2008), we use net worth covenants as our primary measure, and tangible net worth covenants when net worth covenants are not available. In our sample, we only include firms which have not yet violated net worth covenants. We do so because the expected earnings management behavior of firms already in technical default is unclear. 7 Using net worth covenant data from Dealscan between 1990 and 2008, we find that the overall level of real earnings management is positively associated with the tightness of net worth covenant slack. This association exists both before and after the adoption of SOX. We further find that our results are driven by loan-years with the smallest net worth covenant slack, i.e. cases when incentives to avoid violations of debt covenants are particularly strong. Finally, consistent with our conjecture that higher debt covenant renegotiation costs increase firms incentives to engage in earnings management, we find that our results are more pronounced when a borrower experiences an increase in bankruptcy probability (estimated default frequency) in the previous year. We contribute to the growing literature on real earnings management activities. Early work by Bartov (1993) and Haw et al. (1991) and a conjecture in Roychowdhury (2006) provide initial clues that real activities manipulation could be related to the structure of 5 Other common income-based types of covenants reported in Dealscan include debt-to-ebitda, debt-toincome, debt-to-equity, debt-to-tangible net worth, debt-service-coverage, interest coverage, and fixed charge ratio covenants. Because the computation of these covenant types could differ among contracts, our estimates of slack tightness for such covenants are likely to be noisy. Hence, we do not consider them in our analyses. 6 Chava and Roberts (2008) also consider current ratio covenants. However, the impact of real activity manipulation on current ratios is less clear. Hence, we do not use this covenant type in our analyses. 7 For instance, one could argue that, once in technical default, firms have no additional incentives to continue to manage earnings. Alternatively, it could be argued that they still manage earnings anyway to avoid even deeper default. This could be explored in future work. 5

private debt contracts. We provide direct evidence to that effect. Recent work suggests that real earnings management is likely to be more pronounced in the presence of stronger opportunistic incentives to manage earnings, such as higher levels of fixed costs (Gupta, et al. 2010), SEOs (Cohen and Zarowin, 2009), seeking to avoid reporting losses (Roychowdhury, 2006), and generally higher levels of debt (Roychowdhury, 2006). We extend this literature to the private debt contract setting. In doing so, we also provide support for the Debt Covenant Hypothesis advanced by Positive Accounting Theory. Our paper proceeds as follows. Section 2 develops our hypotheses. Section 3 describes our research design and sample. Section 4 discusses results, and Section 5 concludes. 2. Hypotheses development The Debt Covenant Hypothesis posits that managers make accounting choices to avoid debt covenant violations because violating covenants is costly (Watts and Zimmerman, 1986). Chava and Roberts (2008) document that disclosed debt covenant violations result in significant declines in future investments in a firm, as creditors take actions to protect their collateral. This is consistent with the evidence in Nini et al. (2009) showing that debt contracts contain provisions that restrict capital expenditures by borrowers. Roberts and Sufi (2009) show that, following debt covenant violations, firms interest costs increase and the availability of credit decreases. Core and Schrand (1999) find that negative earnings news has a negative valuation effect on thrift institutions close to violation of debt covenants. These results suggest that disclosed debt covenant 6

violations are costly. Hence, firm managers have incentives to engage in earnings management to avoid disclosing such violations. Dichev and Skinner (2002) and Armstrong et al. (2009) point out that disclosed debt covenant violations represent a sub-sample of the most extreme cases of covenant violations, i.e., those which cannot be cured by renegotiation. Indeed, most violations are negotiated or waived, and, therefore, are never announced. However, even if a firm is successful in avoiding announced violations, Roberts and Sufi (2009b) show that debt contract renegotiations are associated with increased likelihood of negative outcomes of renegotiations for borrowers, such as reduced amount of loan principal or increased loan spreads. Moreover, renegotiation of covenants is a costly process that requires payments of legal fees, time and resources, e.g., costs of negotiations with lenders and the involvement of lawyers in renegotiations of actual debt contracts 8. Hence, covenant violations, whether cured or not, are likely to be costly to borrowers. Consistent with this argument, the Debt Covenant Hypothesis predicts that managers will take actions to shield themselves from these negative effects and engage in activities that ex-ante reduce the likelihood of future debt covenant violations (Fields et al., 2001). Consistent with this prediction, prior evidence in the literature shows that in order to avoid debt covenant violations, firms choose more favorable accounting policies such as more favorable depreciation methods, inventory valuation methods (FIFO/LIFO), longer 8 It is a common view in the literature that disclosed covenant violations represent extreme cases where the renegotiation route could not be pursued (Dichev and Skinner, 2002, Armstrong et al. 2009). This is consistent with the SEC s disclosure requirement of only those violations which could not be cured. However, Roberts and Sufi (2009a) find evidence that some of the disclosed violations indeed represent violations which were cured (i.e. successfully renegotiated), i.e. management chooses to voluntarily disclose some violations. Because Roberts and Sufi s primary finding that disclosed violations are associated with substantial increases in the cost of debt, one possible implication of their study is that even cured violations could entail costs far higher than the simple costs of renegotiations we list above. 7

amortization periods for prior period pension costs, and manipulate abnormal accruals. 9 Because prior evidence suggests that conditional conservatism is more likely to trigger violations of debt covenants (Zhang, 2008), managers are also more likely to adopt less conservative accounting policies when debt covenant slack is small (Kim, 2009). In addition to engaging in accrual earnings management to avoid covenant violations, managers could also use real earnings management (Armstrong et al, 2009). Accrual earnings management is likely to be more costly to managers because it exposes firms to auditors scrutiny, class-action securities litigation, SEC investigations, and, in the extreme, exposes managers to criminal liability. 10 On the other hand, real earnings management does not expose firms to auditor scrutiny or legal liability because it does not involve misleading disclosures and intentional manipulation of accounting numbers. In other words, real earnings management might be myopic in nature, but security laws do not penalize managers for sub-optimal decisions. Hence, real earnings management is less costly to managers than accrual earnings management, because managers cannot legally be held responsible for real earnings management as long as the outcomes of real earnings management are properly disclosed in the financial statements. Consistent with this argument, Cohen et al. (2008) show that real earnings management becomes more prevalent than accrual earnings management after adoption of the Sarbanes-Oxley Act.Furthermore, Zang (2007) shows that the switch from accruals earnings management to real earnings management is more likely among firms facing greater litigation risk. 9 See, for example, Holthausen, 1981; DeFond and Jiambalvo, 1994; Sweeney, 1994; Beneish, Press, and Vargus, 2001. 10 See, for example, Roychowdhury 2006, Cohen et al. 2008, Zang 2007, Dechow, Sloan and Sweeney 1996, Dechow, Ge, Larson, and Sloan 2007. 8

These results suggest that, faced with the need to manage earnings, managers could choose real earnings management as a means to avoid debt covenant violations. Prior research finds some initial evidence of the use of real earnings management to avoid debt covenant violations. For instance, Roychowdhury (2006) finds that real earnings management is higher for firms with debt versus. firms without debt. Similarly, Bartov (1993) finds that income from asset sales is higher for firms with higher leverage, with higher leverage serving as a proxy for more restrictive debt covenants. Haw et al. (1991) find some evidence that the presence of public debt covenants and higher leverage are associated with greater likelihood of settling over-funded pension obligations, resulting in partial settlement gain recognition in earnings and the avoidance of debt covenant violations. All of these studies suggest that the avoidance of debt covenant violations could motivate managers to engage in real activities manipulations. However, these studies face the same basic limitation. They use the debt-to-equity ratio or the presence of debt as an indirect proxy of debt covenant slack (Fields et al., 2001). In addition, Haw et al. s (1991) use of the presence of public debt covenants is not a direct measure of proximity to covenant violation. Also, Haw et al. (1991) do not examine the use of real earnings management to avoid violation of private debt covenants. Finally, Bartov (1993) and Haw et al. (1991) focus on only some manifestations of real activities manipulation, such as asset sales, or pension fund settlements. As Roychowdhury (2006) points out, the menu of real activity manipulation choices is potentially much broader because it includes manipulation of inventory over-production decisions, discretionary expenses, and sales margins. Thus, we examine whether firms with tighter net worth covenants are 9

more likely to engage in various real earnings management. Our first hypothesis (in the alternative form) is: H1: Borrowers with tighter covenant slack are more likely to manipulate real activities. Prior literature on earnings management shows that firms are more likely to manage earnings when earnings management is going to be most beneficial, e.g. when it helps firms avoid missing expectations. Roychowdhury (2006) shows that firms close to missing analyst expectations and/or reporting losses have higher levels of real earnings management. Baber et al. (1991) show that firms are more likely to reduce their R&D expenditures when their income is lower. Bushee (1998) shows that transient institutional investor pressure affects managers decision to cut R&D in order to avoid missing earnings targets. Similarly, we expect managers to use real earnings management to a greater extent when their firms are closer to a zero threshold of debt covenant slack. This leads to the following hypothesis: H2: Borrowers are more likely to manipulate real activities when they are closer to the zero covenant slack threshold. Managers are more likely to resort to real earnings management to avoid violations of debt covenants when they expect outcomes of covenant violations to be unfavorable. An unfavorable outcome of renegotiations is more likely when the lender perceives a higher risk of default on the part of the borrower. In this situation, a lender could impose on the borrower less favorable terms (for example, reductions in loan principal or higher interest rate), or in the extreme case a lender could call on the loan altogether. This suggests that 10

a borrower with an increased bankruptcy risk should have stronger incentives to take actions to avoid covenant violations ex-ante instead of attempting to renegotiate ex-post. This reasoning leads to the following prediction: H3: Increased bankruptcy risk will strengthen the positive relation between the tightness of covenant slack and real earnings management for the firms that are close to covenant violations. 3. Research Design and Sample 3.1. Research design To test our hypotheses, we follow the Roychowdhury (2006) and Cohen et al. (2008) models 11 of real earnings management and augment them with our variables of interest that capture tightness of debt covenant slack (T_SLACK), degree of proximity to zero slack (CLOSE), and prior year increase in estimated default frequency ( BANK). In particular, to test H1, we run the following regression model: REM t =a 0 +a 1 *T_SLACK t +a 2 *LMVE t-1 +a 3 *MTB t-1 +a 4 *LEV t-1 +a 5 *ROA t-1 +a 6 * GDP t-1 +a 7 *ABSDA t +e t (1) To test H2, we run the following regression model: REM t =b 0 +b 1 *T_SLACK t + b 2 *T_SLACK t * CLOSE t + b 3 * CLOSE t +b 4 *LMVE t-1 +b 5 *MTB t-1 +b 6 *LEV t-1 +b 7 *ROA t-1 +b 8 * GDP t-1 +b 9 *ABSDA t +e t (2) 11 Cohen et al. s (2008) model also includes a number of executive compensation controls (such as executives levels of salaries, bonus, stock options and restricted stock ownership levels). Because requiring availability of these variables in conjunction with Dealscan data significantly reduces our sample size, we do not use these variables in our main tests. However, we include them in our robustness tests, and our results remain consistent with those reported. 11

To test H3, we run the following regression model: REM t =c 0 +c 1 *T_SLACK t + c 2 *T_SLACK t * CLOSE t + c 3 *T_SLACK t * CLOSE t * BANK t-1 + c 4 *T_SLACK t * BANK t-1 + c 5 * CLOSE t * BANK t-1 +c 6 * CLOSE t + c 7 * BANK t-1 +c 8 *LMVE t-1 +c 9 *MTB t-1 +c 10 *LEV t-1 +c 11 *ROA t-1 +c 12 * GDP t-1 +c 13 *ABSDA t +e t (3) where the variables are defined as follows: REM T_SLACK LMVE MTB LEV ROA GDP ABSDA CLOSE We define real earnings management variables following Roychowdhury (2006): ABN_CFO: Abnormal cash flows (negative measure of real earnings management) ABN_Prod: Abnormal inventory over-production (positive measure of real earnings management) ABN_Discexp: Abnormal discretionary expenses (negative measure of real earnings management) REM_Index: ABN_Prod/std(ABN_Prod) - ABN_CFO/std(ABN_CFO) - ABN_Discexp/Std(ABN_Discexp) where std(.) stands for standard deviation of each respective variable. Tightness of debt covenant slack. We combine net worth and tangible net worth covenant slack into a single net worth covenant slack variable. For net worth covenant, the tightness of covenant slack is defined as the required minimum net worth covenant per DealScan minus actual common equity (CEQ), deflated by prior year s total assets. For tangible net worth covenant, the tightness of covenant slack is defined as the required minimum tangible net worth covenant per DealScan minus actual common equity less intangible assets, deflated by prior year s total assets. A larger (i.e., less negative) T_SLACK indicates tighter covenant, i.e., closer to violation. Natural log of market value of equity for a firm A firm s market-to-book ratio A firm s leverage defined as the ratio of total liabilities to assets A firm s return on assets defined as the ratio of earnings before extraordinary items deflated by prior period assets Change in annual level of US GDP as reported by the US Bureau of Economic Analysis Absolute value of modified Jones (1991) model of discretionary accruals, with control for contemporaneous accounting performance as suggested in Kothari et al (2005). A dummy variable equal to one if the magnitude of T_SLACK is 12

BANK between 0 and -10%, and zero otherwise. A dummy variable equal to one if a firm experiences an increase in estimated default frequency (Merton KMV Measure of bankruptcy risk, estimated using an algorithm in Bharath and Shumway, 2008) in the prior year. Our H1 predicts that real earnings management will increase as T_SLACK increases. Hence, we expect that the coefficient on T_SLACK (a 1 ) in model (1) will be positive for models that use REM_Index and ABN_Prod as dependent variables, and negative for models that use ABN_CFO and ABN_Discexp as dependent variables. H2 predicts that the extent of real earnings management is stronger for firms closest to the debt covenant threshold. Hence, under this hypothesis we expect thatb 2 and b 1 +b 2 in model (2) will be positive for models using REM_Index and ABN_Prod as dependent variables, and b 2 and b 1 +b 2 will be negative for models using ABN_CFO and ABN_Discexp as dependent variables. H3 predicts that the association between tighter debt covenant slack and real earnings management is stronger for firms that experience increases in bankruptcy risk. Hence, under this hypothesis we expect that c 3 and c 1 +c 2 +c 3 +c 4 in model (3) will be positive for models using REM_Index and ABN_Prod as dependent variables, and c 3 and c 1 +c 2 +c 3 +c 4 will be negative for models using ABN_CFO and ABN_Discexp as dependent variables. We follow Roychowdhury (2006) and Cohen et al. (2008) to control for variables (LMVE, MTB, LEV, ROA, GDP, and ABSDA) that affect real earnings management. We also control for fixed-year effects. To mitigate the influence of potential outliers, we winsorize all continuous variables at their respective 1 st and 99 th percentiles. Following Gow et al. (2010), we report test statistics based on the two-way cluster-robust standard 13

errors (cluster by firm and by year), which adjust for both cross-sectional and time-series dependence in panel data. 3.2. Sample and descriptive statistics 3.2.1. Sample selection We obtain a sample of all loans with net worth and tangible net worth debt covenants from Dealscan and match it with Compustat using a matching dataset provided by Michael Roberts and originally used in Chava and Roberts (2008). Chava and Roberts (2008) examine how violations of net worth and current ratio debt covenants impact firms investment policy. 12 Because Dealscan does not have a corresponding matching identifier to Compustat (such as ticker or CUSIP), Michael Roberts provides a matching dataset that can be used to merge Dealscan and Compustat data. To the extent feasible, we also manually match Dealscan and Compustat data for observations not available in Michael Roberts matching dataset. Our sample period starts in 1990 and ends in 2008. We require Compustat data availability of real earnings management measures, as well as prior year size, ROA, leverage, market-to-book, and discretionary accruals measures. This procedure results in a full sample of 6,144 loan-years. We exclude any observations with positive T_SLACK (i.e. negative slack firms, or firms in technical default) from our sample. 13 We exclude these observations for two reasons: 1) technical default firms have already violated their covenants and hence it is unclear whether they have further incentives to manage earnings through real activity manipulations; 2) it is possible that negative debt covenant slack (positive T_SLACK) can be a result of an unobservable 13 The variable T_SLACK represents tightness of slack, i.e., negative of the actual reported slack. 14

subtlety in the definition of the covenant (Chava and Roberts, 2008). 14 This sample selection procedure results in a sample of 4,826 loan-years. We use this sample to estimate models (1) and (2). Model (3) requires availability of lagged change in estimated default frequency, which reduces our sample size to 4,200 loan-year observations. 3.2.2. Descriptive statistics Table 1 presents descriptive statistics for our sample. The means of the amount of real earnings management variables, i.e., abnormal cash flow, abnormal production, and abnormal discretionary expenditures are 0.026, -0.002, and -0.005, respectively. The mean of REM_Index is 0.003 and the median is -0.064. Since we exclude all firms already in technical default, all our firms have negative T_SLACK. The mean of T_SLACK is -0.216 and the median is -0.159. The mean of Lev is 0.609, suggesting that companies finance their assets with 60.9% of debt. Because debt is a significant source of financing for these companies, they will have stronger incentives to avoid covenant violations due to the potential discontinuity of projects from covenant violations. Table 2 presents the pair-wise Pearson correlations. REM_Index is positively correlated with T_SLACK, consistent with firms managing real activities to avoid the net worth debt covenant violations. Furthermore, the correlation table suggests that firms manage each of the three components of real activities, i.e., cash flows, production and discretionary expenditures, to avoid violating the net worth covenant violations. 4. Empirical Results 14 This phenomenon of covenant violations in the quarter of the loan origination is encountered by Dichev and Skinner (2002) and Chava and Roberts (2008). 15

Table 3 presents the regression results. We find a positive coefficient of 1.031 (t=5.50) on T_SLACK in the REM_Index regression, suggesting that firms manage more real activities when their net worth covenant slack is tighter, i.e., they manage real activities to avoid violating net worth covenant violations. With respect to our control variables, we find a negative coefficient on LMVE, a negative coefficient on MTB, a positive coefficient on LEV, and a negative coefficient on ROA. These results are consistent with the findings in the literature (Roychowdhury 2006, Cohen et al. 2008). When we look at each component of real earnings management, we find the coefficient on T_SLACK to be negative (t=-2.22) in the ABN_CFO regression, positive (t=2.57) in the ABN_Prod regression, and negative (t=-5.69) in the ABN_Discexp regression. These results suggest that firms lower abnormal cash flows, increase abnormal production and cut down abnormal discretionary expenditures to avoid violating net worth covenant violations. Overall, our results are consistent with H1. Since H2 predicts that firms have stronger incentives to manage earnings when their covenant slack is the tightest, we test whether firms manage real activities to a larger extent when their net worth covenant slack is close to zero. We present our model (2) results in Table 4. Consistent with H2, we find a positive coefficient of 3.822 (t=1.83) on T_SLACK*CLOSE in the REM_Index regression, suggesting that firms that are very close to violating their net worth covenant are more likely to manage real activities than other firms. We further examine the total effect of T_SLACK on REM_Index for firms with CLOSE=1, i.e., the association between T_SLACK and real earnings management for firms that are very close to violating these covenants. The sum of the coefficient on T_SLACK and T_SLACK*CLOSE is 4.896, which is significantly positive (t=2.29), 16

suggesting that firms that are close to net worth covenant violations are more likely to manage real activities to avoid such violations. When we further examine the components of real activities manipulation, we find that these firms mainly lower their abnormal cash flows and cut back their discretionary expenditures to accomplish their real earnings management goals. Since H3 predicts that bankruptcy risk increases the cost of covenant violations, we expect that our model (2) results are stronger when firms experience an increase in bankruptcy risk in the prior year. We report the regression results of model (3) in Table 5. Consistent with H3, we find a positive coefficient of 8.870 (t=1.98) on T_SLACK*CLOSE* BANK in the REM_Index regression, suggesting that firms that are very close to violating their net worth covenant are more likely to manage real activities when they experience an increase in bankruptcy risk than firms that do not experience an increase in bankruptcy risk. Furthermore, we examine the total effect of T_SLACK on REM_Index for firms with CLOSE=1 and BANK=1, i.e., the association between T_SLACK and real earnings management for firms that have the strongest incentives to avoid these covenant violations (i.e., firms that very close to violate these covenants and experience an increase in bankruptcy risk). The sum of the coefficient on T_SLACK, T_SLACK*CLOSE, T_SLACK*CLOSE* BANK, and T_SLACK* BANK is 8.342, which is significantly positive (t=3.21), suggests that firms very close to net worth covenant violations and experience an increase in bankruptcy risk overall manage real activities to avoid such violations. When we further examine the components of real activities manipulation, we find that these firms accomplish their real earnings management goals 17

mainly through decreasing their abnormal cash flow and increasing their abnormal production. 5. Additional Analyses 5.1. Modeling endogeneity of T_SLACK In our primary analyses, we assume that creditors do not structure debt covenants to fully anticipate real earnings management. This assumption underlies many studies in the Debt Covenant Hypothesis literature. As Fields et al. (2001) argue, evidence exists to support this assumption on the basis of the demonstrated inability of sophisticated intermediaries to anticipate accruals earnings management. 15 Lenders might find it more difficult to predict real earnings management than accruals earnings management or opportunistic accounting choice because real activity manipulations could take multiple forms, all of which could be too difficult to write in a complete contract. Also, when contracts are negotiated, managers are less likely to agree to ex-ante future restrictions in real activities because such activities could be value-maximizing business decisions. 16 In contrast, accounting choices come from a limited menu (such as LIFO vs. FIFO), and lenders could theoretically limit borrowers more easily to a certain set of accounting choices ex-ante 17. Lenders will have a harder time doing so with ubiquitous real activity manipulations. 15 Fields et al. (2001) cite findings that sophisticated intermediaries, such as analysts, cannot see through accounting choices. That is, development of sophisticated statistical models to detect earnings management in general is costly, and lenders may lack sophisticated knowledge to do it (see page 289 of Fields et al., 2001). 16 For example, it is hard for lenders to ex-ante restrict managers from future R&D cuts or from decisions to over-produce because ex-ante the optimal levels for such decisions are unknown. 17 Beatty et al. (2002) find that the lenders offer lower interest rates by excluding voluntary accounting changes in the contracts. 18

Nonetheless, to mitigate additional endogeneity concerns, we employ a two-stage procedure similar to Nikolaev (2010), which models the endogeneity of debt covenants in public debt contracts. The first stage model estimates abnormal tightness of net worth covenant slack (ABN_T_SLACK). Roychowdhury (2006) shows that real earnings management is more pronounced among firms that experience lower levels of accounting performance, have more long term debt and more current liabilities, are smaller, and have lower growth options. Manufacturing firms are also more likely to resort to real earnings management. The presence of these firm characteristics could lead lenders to set tighter initial slack. In addition, El-Gazzar and Patena (1991) show that the slack is a function of the number of covenants present in a particular debt contract. Nikolaev (2010) also shows that the degree of restrictiveness of public debt covenants varies with changes in firm leverage, asset tangibility (ratio of fixed assets to total assets), 18 dividend yield, credit rating of the borrower, Altman Z-score, and remaining maturity and amount of the loan. We use all of these variables to construct the first stage model: 1 st stage equation: T_SLACK t =a 0 +a 1 * LMVE t-1 +a 2 * BTM t-1 +a 3 *LEV t-1 + a 4 *ROA t-1 + a 5 *Loss t +a 6 * LEV t-1 +a 7 *DIV_YIELD t-1 +a 8 *TANGIBILITY t-1 +a 9 * Z_Score t-1 + a 10 *CL t + a 11 *Log(Maturity) t +a 12 * NCR t + a 13 *MFG t + a 14 *N_ Covenants t + a 15 *Log(Amount) t +a 16 *NOA t-1 +e t (4) In this model, BTM is book-to-market ratio, LEV is change in overall leverage, DIV_YIELD is the dividend yield of a firm (ratio of total dividends paid to stock price), TANGIBILITY is the ratio of net fixed assets to total assets, Z_Score is Altman Z-score, CL is the ratio of current liabilities to total assets, Log(Maturity) is natural logarithm of 18 See also, Gupta, Pevzner and Seethamraju (2010) who show that inventory over-production is affected by the ratio of fixed assets to total assets. 19

the remaining maturity of the loan, NCR is a dummy variable equal to 1 if a firm does not have a long-term credit rating in Compustat, MFG is a dummy variable for manufacturing firms (SIC codes between 2000 and 3999), N_Covenants is the total number of covenants in a particular loan, Log(Amount) is natural logarithm of the amount of the loan, and NOA is the level of a firm s net operating assets deflated by prior year assets (used to control for earnings management constraint per Barton and Simko, 2002 and Cohen and Zarowin, 2009). All other variables are as defined earlier. We obtain the residuals e t (ABN_T_SLACK ) from the first-stage model (4). To control for endogeneity, in the second stage, we replace T_SLACK with ABN_T_SLACK in our regression model (1) following the methodology in Nikolaev (2010). 19 Intuitively, one can think of ABN_T_SLACK as an instrumental variable for T_SLACK; while it is correlated with T_SLACK, by construction, it is uncorrelated with other variables that endogenously determine T_SLACK. We report the two-stage results in Table 6. 20 Panel A reports our first stage results and Panel B report our second stage results for model (1). Consistent with H1 and our Table 3 results, we continue to find positive coefficients on ABN_T_SLACK in the REM_Index (t=4.61) and ABN_Prod (t=2.06) regressions, and 19 We also conduct three additional analyses to address the endogeneity of T_SLACK. First, we follow Carcello et al. (2010) and add ABN_T_SLACK as an additional regressor in our model (1). We continue to find significant coefficients on T_SLACK in all four regressions, all signs consistent with our predictions in H1. Second, we again follow Carcello et al. (2010) and use Stata s SUEST command to simultaneously estimate Model (1) and Model (4). Again, we continue to find significant coefficients on T_SLACK in all four regressions, and the signs are consistent with our predictions in H1. Second, we use Stata s IVREGRESS command to estimate Model (1) where the first-stage model is specified as in Model (4). We continue to find significant coefficients on T_SLACK in all four regressions, all signs consistent with our predictions in H1. In summary, our results are robust to controlling for endogeneity of the tightness of net covenant slack. We do not tabulate these additional endogeneity analyses results for brevity. 20 We do not report the results of Models (2) and Model (3) after controlling for endogeneity of T_SLACK because there is no econometrically sound way to correctly address the endogeneity issue when the endogenously determined variable (T_SLACK) appears in two-way (e.g., T_SLACK*CLOSE) and threeway (T_SLACK*CLOSE* BANK) interaction terms. To add to the complexity, our CLOSE variable is an indicator variable defined based on T_SLACK, which is endogenously determined. Thus, it is empirically very difficult, if not impossible, to correctly control for endogeneity in Models (2) and (3). 20

negative coefficients in the ABN_CFO (t=-2.70) and ABN_Discexp (t=-2.82) regressions. These findings suggest that our conclusion that tighter net worth covenant slack are associated with higher levels of real earnings management is robust to controlling for endogeneity of the tightness of slack. 5.2. Pre and post-sox analyses Cohen, Dey and Lys (2008) show that real earnings management increased after the adoption of the Sarbanes-Oxley Act (SOX). Their findings suggest that the association between the tightness of the net worth covenant slack and real earnings management could be more prominent post-sox. To test this conjecture, we re-run models (1) after including the interaction of T_SLACK with SOX, where SOX=1 if a firm has a fiscal year after 2004 and 0 if it is before 2003. We exclude 2003 from our analysis since it was a transitional period for SOX. We also include the main effect of SOX in the model. We exclude the year dummies because those are redundant with the SOX dummy. Our results (untabulated for brevity) show that the interaction T_SLACK*SOX is not significant except in the ABN_Prod regression, while the main effect on T_SLACK stays consistent with those reported in Tables 3-5. Untabulated tests regarding the total effect of T_SLACK on real earnings management variables (the sum of the coefficients on T_SLACK and SOX*T_SLACK) suggest that firms manage the overall level of real activities through all three types (i.e., abnormal cash flow, abnormal production, and abnormal discretionary expenditures) to avoid violating net worth covenant post-sox. Overall, these results suggest that firms used real activities manipulation to avoid debt covenant violations both before and after SOX adoption, and that SOX largely did not 21

affect their propensity to do so, possibly because incentives to avoid debt covenant violations are strong both before and after SOX. 5.3. Robustness checks Following Cohen et al. (2008), we also control for the level of unexercised stock options, exercised stock options, managerial bonus and salary obtained from ExecuComp. Our results remain similar to those reported.furthermore, we replace the absolute value of discretionary accruals with signed discretionary accruals in our models (1)-(3) and the results are robust. 5. Conclusions We examine whether tighter net worth covenant slack is associated with greater levels of real earnings management. Using actual net worth debt covenant data from Dealscan, we find evidence consistent with this hypothesis. Moreover, we find that this result is more pronounced for loan years with the tightest slack, and within that sub-sample, the result is driven by firms experiencing increases in their bankruptcy risk in the prior year. Our results continue to hold after controlling for the endogeneity of debt covenant slack. Hence, our findings provide further support for the Debt Covenant Hypothesis suggesting that managers will take real actions to avoid potentially costly debt covenant violations. Prior work finds that real earnings management is higher for firms with more debt (Bartov, 1993, Roychowdhury, 2006), and that some real earnings management activities are more pronounced in the presence of public debt covenants (Haw et al. 1991). Prior studies show that less conservative accounting choices and accrual earnings management 22

are more pronounced when debt covenant slack is tighter (DeFond and Jiambalvo, 1994, Beneish, Press and Vargus, 2001, Kim 2009). Our study is the first to provide direct evidence that the tightness of private debt covenant slack also affects real activity manipulations. We also provide additional support to prior literature that finds that real earnings management is driven by incentives to manipulate earnings, such as avoiding missing earnings targets (Roychowdhury, 2006) or maximizing proceeds from SEOs (Cohen and Zarowin, 2009). Our results suggest that a tighter debt covenant is another context in which real activities manipulation is likely to occur. 23

References: Armstrong, C., W. Guay, J. Weber (2009). The role of information and financial reporting in corporate governance and contracting. Forthcoming, Journal of Accounting and Economics. Barton, J. and P. Simko (2002). Balance sheet as an earnings management constraint. The Accounting Review, 77, Supplement, 1-27. Baber, W., P. Fairfield, and J. Haggard (1991). The effect of concern about reported income on discretionary spending decisions: the case of research and development. The Accounting Review, 66(4), 818-829. Bartov, E., (1993). The timing of asset sales and earnings manipulation. The Accounting Review 68, 840 855. Beatty, A., Ramesh, K., and Weber, J. (2002). The importance of accounting changes in debt contracts: the cost of flexibility in covenant calculations. Journal of Accounting and Economics, 33, 205-227. Beneish, M., Press, E., Vargus, M., 2001. The relation between incentives to avoid debtcovenant default and insider trading. Working paper, Indiana University, Temple University, and University of Texas at Dallas. Bharath, S. and T. Shumway (2008). Forecasting Default with the Merton Distance to Default Model. Review of Financial Studies, 21(3), 1339-1369. Bushee, B. (1998). The influence of institutional holders on myopic R & D investment behavior. The Accounting Review, 73(3), 305-333. Carcello, J. V., T. L. Neal, Z-V. Palmrose, and S. Scholz (2010). CEO Involvement in Selecting Board Members, Audit Committee Effectiveness, and Restatements. Contemporary Accounting Research (forthcoming). Chava, R. and M. Roberts (2008). How does financing impact investment? The role of debt covenants. Journal of Finance, 63, 2085-2121. Cohen, D., A. Dey, T. Lys (2008). Real and accrual-based earnings management in the pre and post-sarbanes-oxley periods. The Accounting Review, 82(3), 757-787. Cohen, D. and P. Zarowin (2009). Earnings management and excess investment: Accrual-based versus Real Activities. Working Paper. New York University. Cohen, D. and P. Zarowin (2010). Accrual based and real earnings management activities 24

around seasoned equity offerings. Journal of Accounting and Economics, forthcoming. Core, J. and C. Schrand (1999). The effect of accounting-based debt covenants on equity valuation. Journal of Accounting and Economics, 27(1), 1-34. Dechow, P., R. Sloan, and A. Sweeney (1996). Causes and Consequences of Earnings Manipulation: An Analysis of Firms Subject to Enforcement Actions by the SEC. Contemporary Accounting Research, 13(2). Dechow, P., W. Ge, C. Larson, and R. Sloan (2007). Predicting material accounting manipulations. Working Paper. University of Michigan. DeFond, M. and J. Jiambalvo (1994). Debt Covenant Violation and Manipulation of Accruals. Journal of Accounting and Economics 17, 145-76. Dichev, I. and D. Skinner (2002). Large sample evidence on debt covenant hypothesis. Journal of Accounting Research, 40(4), 1091-1123. El-Gazzar, S. and V. Pastena (1991). Factors affecting the scope and initial tightness of covenant restrictions in private lending agreements. Contemporary Accounting Research, 8(1), 132-151 Fields, T., T. Lys, and L. Vincent (2001). Empirical research on accounting choice. Journal of Accounting and Economics, 31(1), 255-307. Gow, I. D., G. Ormazabal, and D. J. Taylor (2010). Correcting for Cross-Sectional and Time-Series Dependence in Accounting Research. The Accounting Review 85 (2), 483-512. Graham, J.R., Harvey, C.R., Rajgopal, S.(2005). The economic implications of corporate financial reporting. Journal of Accounting and Economics 40, 3 73. Gupta, M., M. Pevzner, and C. Seethamraju (2010). The implications of absorption cost accounting and production decisions for future firm performance and valuation. Working Paper. Washington University. Haw, I, K. Jung,, and S. Lilien (1991). Overfunded defined benefit pension plan settlements without asset reversion. Journal of Accounting and Economics 14, 295 320. Holthausen, R., (1981). Evidence on the effect of bond covenants and management compensation contracts on the choice of accounting techniques. Journal of Accounting and Economics 3, 73-109. Jones, J. (1991). Earnings management during import relief investigations. Journal of 25

Accounting Research, 29, 193-228. Kim, B. (2009). Post-borrowing conservatism and debt covenant slack. Working Paper. American University. Kothari, SP, A. Leone, and C. Wasley (2005). Performance-matched discretionary accruals measures. Journal of Accounting and Economics, 39,163-197. Nini, G, D. Smith, and A.Sufi (2009). Creditor control rights and firms investment policy. Journal of Financial Economics, 92(3), 400-420. Nikolaev, V. (2010). Debt covenants and accounting conservatism. Journal of Accounting Research, 48(1), 137-175. Roberts, M. and A. Sufi (2009a). Control rights and Capital Structure: An Empirical Investigation. Journal of Finance, 64, 1657-1695. (2009b). Renegotiation of financial contracts: Evidence from private credit agreements. Journal of Financial Economics, 93,159-184. Roychowdhury, S. (2006). Earnings management through real activities manipulation. Journal of Accounting and Economics, 42(3), 335-370. Sweeney, A., (1994). Debt-covenant violations and managers accounting responses. Journal of Accounting and Economics 17, 281 308. Watts, R. and J. Zimmerman (1986). Positive Accounting Theory, Prentice Hall. Zang, A. (2007). Evidence on the Tradeoff Between Real Manipulation and Accrual Manipulation. Working Paper. HKUST. Zhang, X. (2007). Economic consequences of the Sarbanes Oxley Act of 2002. Journal of Accounting and Economics, 44(1), 74-115. Zhang, J. (2008). The contracting benefits of accounting conservatism to lenders and borrowers. Journal of Accounting and Economics, 45(1), 27-54. 26

Appendix: Variables Definitions REM Real earnings management variables following Roychowdhury (2006): ABN_CFO: Abnormal cash flows (negative measure of real earnings management) ABN_Prod: Abnormal inventory over-production (positive measure of real earnings management) ABN_Discexp: Abnormal discretionary expenses (negative measure of real earnings management) REM_Index: ABN_Prod/std(ABN_Prod) - ABN_CFO/std(ABN_CFO) - ABN_Discexp/Std(ABN_Discexp) where std(.) stands for standard deviation of each respective variable. T_SLACK Tightness of debt covenant slack. We combine net worth and tangible net worth covenant slack into a single net worth covenant slack variable. For net worth covenant, the tightness of covenant slack is defined as the required minimum net worth covenant per DealScan minus actual common equity (CEQ), deflated by prior year s total assets. For tangible net worth covenant, the tightness of covenant slack is defined as the required minimum tangible net worth covenant per DealScan minus actual common equity less intangible assets, deflated by prior year s total assets. A larger (i.e., less negative) T_SLACK indicates tighter covenant, i.e., closer to violation LMVE Natural log of market value of equity for a firm MTB A firm s market-to-book ratio LEV A firm s leverage defined as the ratio of total liabilities to assets ROA A firm s return on assets defined as the ratio of earnings before extraordinary items deflated by prior period assets GDP Change in annual level of US GDP as reported by the US Bureau of Economic Analysis ABSDA Absolute value of modified Jones (1991) model of discretionary accruals, with control for contemporaneous accounting performance as suggested in Kothari et al (2005). CLOSE A dummy variable equal to one if the magnitude of T_SLACK is between 0 and -10%, and zero otherwise. BANK A dummy variable equal to one if a firm experiences an increase in estimated default frequency (Merton KMV Measure of bankruptcy risk, estimated using an algorithm in Bharath and Shumway (2008)) in the prior year. BTM A firm s book-to-market ratio LEV Change in overall leverage DIV_YIELD Dividend yield of a firm (ratio of total dividends paid to stock price) TANGIBILITY Ratio of net fixed assets to total assets Z_Score Altman Z-score CL NCR Ratio of current liabilities to total assets Dummy variable equal 1 if a firm does not have a long-term credit rating in Compustat, and 0 otherwise 27

MFG N_Covenants NOA Log(Maturity) Log(Amount) dummy variable equal to 1 for manufacturing firms (SIC codes between 2000 and 3999), and 0 otherwise Total number of covenants in a particular loan Level of a firm s net operating assets, deflated by prior year assets (used to control for earnings management constraint per Barton and Simko, 2002 and Cohen and Zarowin, 2009) Natural logarithm of the remaining maturity of the loan in months obtained from Dealscan Natual logarithm of the amount of the loan obtained from Dealscan 28

Table 1 Descriptive Statistics This Table summarizes descriptive statistics for variables defined in the Appendix. N Mean Std. Dev. Q1 Median Q3 ABN_CFO t 4826 0.026 0.111-0.029 0.03 0.088 ABN_Prod t 4826-0.002 0.215-0.119-0.015 0.095 ABN_Discexp t 4826-0.005 0.236-0.111-0.01 0.097 REM_Index t 4826 0.003 2.087-1.072-0.064 0.961 T-SLACK t 4826-0.216 0.201-0.281-0.159-0.082 LMVE t-1 4826 5.55 1.695 4.387 5.61 6.674 MTB t-1 4826 2.407 2.322 1.18 1.852 2.841 LEV t-1 4826 0.609 0.354 0.405 0.562 0.717 ROA t-1 4826 0.046 0.109 0.015 0.052 0.092 GDP t-1 4826 3.31 1.216 2.5 3.7 4.4 ABSDA t 4826 0.064 0.064 0.02 0.046 0.087 CLOSE t 4826 0.311 0.463 0 0 1 BANK t-1 4200 0.496 0.5 0 0 1 29

Table 2 Correlations This Table summarizes the pair-wise Pearson correlations for variables defined in the Appendix. * denotes significance level of 10%. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) ABN_CFOt(1) 1 4826 ABN_Prodt(2) -0.3501* 1 4826 4826 ABN_Discexpt(3) -0.0303* -0.7229* 1 4826 4826 4826 REM_Indext(4) -0.5623* 0.9096* -0.7431* 1 4826 4826 4826 4826 T_SLACKt(5) -0.1217* 0.1007* -0.1509* 0.1655* 1 4826 4826 4826 4826 4826 LMVEt-1(6) 0.1633* -0.0939* 0.0096-0.1162* -0.1532* 1 4826 4826 4826 4826 4826 4826 MTBt-1(7) 0.1097* -0.1577* 0.1777* -0.1963* -0.2678* 0.3359* 1 4826 4826 4826 4826 4826 4826 4826 LEVt-1(8) -0.0824* 0.1057* -0.0723* 0.1037* 0.1538* 0.0105 0.0726* 1 4826 4826 4826 4826 4826 4826 4826 4826 ROAt-1(9) 0.2739* -0.0471* -0.1033* -0.0988* -0.1937* 0.2860* 0.1315* 0.0086 1 4826 4826 4826 4826 4826 4826 4826 4826 4826 GDPt-1(10) -0.0142-0.0088 0.0244* -0.0124 0.0167-0.0776* 0.0489* 0.1083* 0.0198 1 4826 4826 4826 4826 4826 4826 4826 4826 4826 4826 ABSDAt(11) -0.0199 0.0283* 0.0770* -0.0055-0.1590* -0.1376* 0.1564* -0.0185-0.0893* 0.015 1 4826 4826 4826 4826 4826 4826 4826 4826 4826 4826 4826 CLOSEt(12) -0.1209* 0.0489* -0.0429* 0.0921* 0.5453* -0.2077* -0.1429* 0.1311* -0.1968* 0.0372* -0.0371* 1 4826 4826 4826 4826 4826 4826 4826 4826 4826 4826 4826 4826 BANKt-1(13) -0.0469* 0.0114-0.0049 0.0328* 0.0973* -0.0851* -0.0979* 0.1238* -0.1302* 0.0465* 0.0199 0.0787* 1 4200 4200 4200 4200 4200 4200 4200 4200 4200 4200 4200 4200 4200 30

Table 3 Regression Analysis This Table summarizes the regressions of real earnings management variables on the tightness of net worth/tangible net worth covenant slack and other control variables (Model 1). All continuous variables are winsorized at the 1 st and 99 th percentiles, respectively. All models include year fixed effects and the t-statistics in parentheses are based on the two-way cluster-robust standard errors (cluster by firm and by year), which adjust for both cross-sectional and time-series dependence in panel data. *, **, *** denote significance levels of 10%, 5%, and 1%, respectively. All variables are defined in the Appendix. REM_Index ABN_CFO ABN_Prod ABN_Discexp Intercept 1.391 *** -0.072 *** 0.077 *** -0.432 *** (7.10) (-6.57) (3.70) (-17.07) T_SLACK t 1.031 *** -0.021 ** 0.051 ** -0.148 *** (5.50) (-2.22) (2.57) (-5.69) LMVE t-1-0.057 * 0.005 *** -0.005 * -0.001 (-1.75) (3.64) (-1.68) (-0.30) MTB t-1-0.145 *** 0.002 ** -0.014 *** 0.017 *** (-7.77) (2.02) (-7.08) (6.41) LEV t-1 0.633 *** -0.025 *** 0.072 *** -0.048 *** (5.93) (-6.05) (5.52) (-2.93) ROA t-1-0.905 ** 0.246 *** -0.011-0.310 *** (-2.26) (7.42) (-0.32) (-6.95) GDP t-1-0.014 0.004-0.002-0.001 (-0.29) (0.94) (-0.42) (-0.13) ABSDA t 0.988-0.009 0.194 *** 0.043 (1.23) (-0.13) (3.41) (0.93) Year Dummies Yes Yes Yes Yes N 4826 4826 4826 4826 Adj. R 2 0.068 0.096 0.045 0.072 31

Table 4 Closeness to Net Worth Covenant Violation This Table summarizes the regressions of real earnings management variables on the tightness of net worth/tangible net worth covenant slack interacting with closeness to net worth covenant violation dummy variable and other control variables (Model 2). All continuous variables are winsorized at the 1 st and 99 th percentiles, respectively. All models include year fixed effects and the t-statistics in parentheses are based on the two-way cluster-robust standard errors (cluster by firm and by year), which adjust for both cross-sectional and time-series dependence in panel data. *, **, *** denote significance levels of 10%, 5%, and 1%, respectively. All variables are defined in the Appendix. REM_Index ABN_CFO ABN_Prod ABN_Discexp Intercept 1.391 *** -0.068 *** 0.078 *** -0.435 *** (6.74) (-6.12) (3.53) (-16.83) T_SLACK t 1.074 *** -0.011 0.061 *** -0.165 *** (4.79) (-1.01) (2.63) (-5.62) T_SLACK t *CLOSE t 3.822 * -0.193 ** 0.334-0.436 * (1.83) (-2.02) (1.33) (-1.91) CLOSE t 0.147-0.018 *** 0.008-0.007 (1.02) (-2.81) (0.43) (-0.39) LMVE t-1-0.057 * 0.005 *** -0.005 * -0.001 (-1.73) (3.37) (-1.71) (-0.22) MTB t-1-0.145 *** 0.002 ** -0.014 *** 0.017 *** (-7.91) (2.11) (-7.29) (6.45) LEV t-1 0.633 *** -0.024 *** 0.072 *** -0.049 *** (6.00) (-5.77) (5.61) (-2.96) ROA t-1-0.897 ** 0.243 *** -0.012-0.308 *** (-2.27) (7.29) (-0.34) (-6.87) GDP t-1-0.009 0.004-0.002-0.002 (-0.18) (0.90) (-0.32) (-0.20) ABSDA t 1.001-0.008 0.196 *** 0.040 (1.24) (-0.12) (3.42) (0.87) Year Dummies Yes Yes Yes Yes N 4826 4826 4826 4826 Adj. R 2 0.069 0.097 0.046 0.073 32

Table 5 Closeness to Net Worth Covenant Violation and Increase in Bankruptcy Risk This Table summarizes the regressions of real earnings management variables on the three-way interaction term (the tightness of net worth/tangible net worth covenant slack, closeness to net worth covenant violation dummy variable, and increase in bankruptcy risk dummy) and other control variables (Model 3). All continuous variables are winsorized at the 1 st and 99 th percentiles, respectively. All models include year fixed effects and the t-statistics in parentheses are based on the two-way cluster-robust standard errors (cluster by firm and by year), which adjust for both cross-sectional and time-series dependence in panel data. *, **, *** denote significance levels of 10%, 5%, and 1%, respectively. All variables are defined in the Appendix. REM_In ABN_CFO ABN_Prod ABN_Discexp dex Intercept 1.412 *** -0.072 *** 0.074 *** -0.436 *** (5.39) (-4.45) (2.88) (-12.72) T_SLACK t 1.561 *** -0.041 ** 0.087 *** -0.185 *** (5.43) (-2.35) (4.23) (-4.29) T_SLACK t *CLOSE t -0.641 0.007-0.188-0.194 (-0.16) (0.04) (-0.40) (-0.50) T_SLACK t *CLOSE t * BANK t-1 8.870 ** -0.458 ** 0.952 ** -0.323 (1.98) (-2.01) (1.98) (-0.75) CLOSE t -0.114-0.003-0.019 0.001 (-0.47) (-0.25) (-0.71) (0.06) BANK t-1-0.346 *** 0.025 *** -0.016 0.010 (-3.26) (3.52) (-1.35) (0.49) T_SLACK t * BANK t-1-1.448 *** 0.090 *** -0.057 * 0.067 (-3.37) (3.75) (-1.70) (1.13) CLOSE t * BANK t-1 0.509 ** -0.036 *** 0.041 * -0.004 (2.11) (-2.86) (1.77) (-0.17) LMVE t-1-0.050 0.005 *** -0.005-0.002 (-1.39) (3.21) (-1.24) (-0.44) MTB t-1-0.138 *** 0.002 * -0.013 *** 0.017 *** (-6.65) (1.80) (-6.02) (6.25) LEV t-1 0.671 *** -0.030 *** 0.074 *** -0.046 *** (5.21) (-4.79) (5.59) (-2.66) ROA t-1-0.590 0.224 *** -0.001-0.315 *** (-1.08) (5.64) (-0.03) (-8.12) GDP t-1-0.016 0.005-0.002-0.001 (-0.26) (0.79) (-0.22) (-0.07) ABSDA t 0.744 0.033 0.221 *** -0.015 (0.79) (0.39) (3.03) (-0.24) Year Dummies Yes Yes Yes Yes N 4200 4200 4200 4200 Adj. R 2 0.059 0.083 0.039 0.066 33

Table 6 Controlling for Endogeneity of T-Slack This Table summarizes the regressions of real earnings management variables on the tightness of net worth covenant slack and other control variables after controlling for the endogeneity of the tightness of net worth covenant slack. All continuous variables are winsorized at the 1 st and 99 th percentiles, respectively. t-statistics in parentheses are based on the two-way cluster-robust standard errors (cluster by firm and by year), which adjust for both crosssectional and time-series dependence in panel data. *, **, *** denote significance levels of 10%, 5%, and 1%, respectively. All variables are defined in the Appendix. Panel A: First Stage Model: Predicting T_SLACK t Coefficient t Intercept -0.602 *** -8.39 LMVE t-1-0.035 *** -6.67 BTM t-1 0.019 * 1.84 LEV t-1 0.079 ** 2.24 ROA t-1-0.082-1.25 Loss t -0.008-0.91 LEV t-1 0.004 0.10 DIV YIELD t-1 0.008 *** 3.07 TANGIBILITY t-1-0.027 * -1.70 Z-Score t-1-0.014 *** -5.20 CL t -0.054 ** -2.40 Log(Maturity) t -0.030 *** -3.42 NCR t 0.004 0.49 MFG t 0.010 1.04 N_ Covenants t -0.001-0.16 Log(Amount) t 0.043 *** 9.12 NOA t-1-0.068 *** -3.76 N 4596 Adj. R 2 0.221 34

Panel B: Second Stage Model (1) REM_Index ABN_CFO ABN_Prod ABN_Discexp Intercept 1.019 *** -0.058 *** 0.059 *** -0.389 *** (5.14) (-5.57) (3.01) (-15.56) ABN_T_SLACK t 0.724 *** -0.031 *** 0.035 ** -0.081 *** (4.61) (-2.70) (2.06) (-2.82) LMVE t-1-0.063 * 0.005 *** -0.005 * -0.000 (-1.93) (3.67) (-1.67) (-0.05) MTB t-1-0.167 *** 0.003 ** -0.015 *** 0.020 *** (-9.45) (2.28) (-8.02) (7.46) LEV t-1 0.811 *** -0.030 *** 0.083 *** -0.068 *** (7.02) (-6.65) (6.33) (-3.69) ROA t-1-1.116 *** 0.251 *** -0.023-0.272 *** (-3.02) (7.47) (-0.64) (-5.53) GDP t-1 0.004 0.002-0.003-0.002 (0.08) (0.41) (-0.56) (-0.23) ABSDA t 0.899-0.000 0.196 *** 0.059 (1.05) (-0.01) (3.21) (1.15) Year Dummies Yes Yes Yes Yes N 4596 4596 4596 4596 Adj. R 2 0.074 0.099 0.045 0.039 35